Cities 64 (2017) 9–17

Contents lists available at ScienceDirect

Cities

journal homepage: www.elsevier.com/locate/cities

Predicting innovative growth and demand with proximate human capital: A case study of the metropolitan area

Juho Kiuru a,⁎, Tommi Inkinen b a , Department of Geosciences and Geography, Division of Urban Geography and Regional Studies, P.O. Box 64, (Gustaf Hällströmin katu 2a), FI-00014 University of Helsinki, b University of Turku, Centre for Maritime Studies, Brahea-Centre, FI-20014 University of Turku, Finland article info abstract

Article history: Human capital is an essential driver for the growth of national and regional innovation systems. In this study, we Received 17 August 2016 can show that also intra-metropolitan innovation clusters locate in, or in proximity to, neighbourhoods with a Received in revised form 4 January 2017 high level of human capital. Our interpretation of human capital involves an educated, talented, creative and Accepted 16 January 2017 tolerant workforce. Indicators from earlier literature are complemented by identified new propositions. In addi- Available online xxxx tion, by using both relative and absolute measures, we conclude that innovations emerge the best in dense and mixed urban structure. After identifying the geography of human capital and innovativeness, we predict also Keywords: Urban planning potential growth areas, which could help urban planning of the HMA. The modelling methods used in this Urban geography study can be implemented and applied in urban studies of other city regions. Human capital © 2017 Elsevier Ltd. All rights reserved. Innovation Predictive analytics

1. Introduction World Economic Forum, 2015). This article presents a statistical analysis of the intra-regional characteristics and interdependencies within the Numerous studies have proven that one of the key drivers for HMA. It will also produce applicable research results for strategic economic growth is human capital. This refers to workforce characteris- planning. We will conduct a predictive analysis on the potential growth tics, including indicators such as the level of education (tertiary), locations based on postal code data. employment knowledge intensity, and employee know-how (e.g. Our research questions are: Glaeser & Saiz, 2004; Alquézar Sabadie & Johansen, 2010). Other studies have applied the concept of creative class, including indicators of talent 1) Do intra-metropolitan clusters of innovation locate in proximity to and tolerance of the workforce (Florida, 2002, 2012). Some studies have neighbourhoods with a high level of human capital? recognised the correlation between human capital and innovativeness 2) How can we predict potential growth areas to help the urban plan- in the national context (Barro, 1991; Rauch, 1993; Simon & Nardinelli, ning of the area? 1996; Simon, 1998). Similar approaches have also been used in regional These research problems relate to both the international context and and urban studies (Zucker, Darby, & Brewer, 1997; Glaeser, 2000; Henry local urban planning. By discovering which indicators are relevant to & Pinch, 2000; Florida, 2002; Florida, Mellander, & Stolarick, 2008; intra-metropolitan innovation clusters, and which indicators of human Lawton Smith, 2009; Boschma & Fritch, 2009; Doms, Lewis, & Robb, capital predict local scale innovativeness, this study contributes to the 2010; Glaeser, Kerr, & Ponzetto, 2010). However, the connection international debate concerning innovation clustering, and especially between human capital (individual capabilities) and spatial innovation to human capital driven growth. clusters (firm properties) has not been greatly studied on an intra- The main findings of our study visualise and statistically verify that metropolitan scale. intra-metropolitan clustering of innovation and human capital hotspots Our study uses the Helsinki Metropolitan Area (HMA) that is the are closely related, not only statistically but also spatially. Our study also capital area and the most important economic concentration in provides indications of areas that could benefit from proximity to these Finland (Makkonen & Inkinen, 2015) as the case location. Finland has innovation hotspots. Another main result is that our analysis illustrates been deemed one of the most innovative countries in the world (e.g. (with solid, diverse, and independently collected datasets) degrees and volumes of distribution in terms of innovation and human capital indi- cators (spatial clustering presented in Fig. 1). In terms of practical sug- ⁎ Corresponding author. gestions, our spatial lag model results provide insights into new E-mail addresses: juho.kiuru@helsinki.fi (J. Kiuru), tommi.inkinen@utu.fi (T. Inkinen). development actions, and bring forth new areas that could have

http://dx.doi.org/10.1016/j.cities.2017.01.005 0264-2751/© 2017 Elsevier Ltd. All rights reserved. 10 J. Kiuru, T. InkinenCities 64 (2017) 9–17

Fig. 1. Geographical distribution of human capital and innovation scores in the HMA. potential for intra-urban strategic planning. The models used in this Education has been widely studied in urban and regional studies, and study can also be used and implemented in empirical urban research it is deemed to be one of the most important societal indicators in other city regions that have similar data resources available. depicting potential and actualised levels of innovation at a location (e.g. Doms et al., 2010; Glaeser et al., 2010; Makkonen & Inkinen, 2. Literature based background 2013). Additionally, the workforce age structure is significant. This re- fers to the number of skilled workers in the age groups close to Human capital has traditionally been considered important in the 40 years (e.g. Bönte, Falck, & Heblich, 2009; Glaeser & Kerr, 2009). (innovative) economic growth of nations and regions, including the Other recognised measures include the number of creative occupations widely applied Porter's diamond approach (e.g. Barro, 1991; Rauch, (e.g. Florida, 2002; Marlet & Van Woerkens, 2004; Mellander & Florida, 1993; Simon & Nardinelli, 1996; Simon, 1998; Porter, 2000) and regions 2006; Boschma & Fritch, 2009; Florida, 2012) and the proportion of (e.g. Zucker et al., 1997; Simon, 1998; Glaeser, 2000; Henry & Pinch, skilled professional immigrant workforce (Stephan & Levin, 2001; 2000; Florida, 2002; Simon & Nardinelli, 2002; Glaeser & Saiz, 2004; Florida, 2002). In this study, we will measure creative occupations Florida et al., 2008; Lawton Smith, 2009; Boschma & Fritch, 2009; with the Standard Industrial Classification (SIC). Doms et al., 2010; Glaeser et al., 2010; Florida, 2012). Fu (2007) demon- From single indicators, inadequate figures for artists, immigrants strated that human capital externalities work locally, even at the census and gay people have been criticised in several studies (e.g. Clark, block level in the Boston Metropolitan Area. In addition, the knowledge 2003; Glaeser, 2004). In this study, different indicators of human capital, spill overs were evident only in the most central census blocks. Fu including the aspect of tolerance, are tested. Instead of the number of named the concept Smart Café Cities. However, in Fu's study, spillover immigrants, we measure areal tolerance, using the popularity of effects were tested between workforce features. Our goal is to deepen immigration-critical political parties as a gauge. We believe that a low the spatial focus of the relationship between human capital and innova- popularity of immigration-critical political parties sufficiently reflects tion. This will also help to bridge the human capital and innovation clus- the tolerance of different areas. We bring a new indicator of workforce tering from regional studies towards urban studies. tolerance into the analysis of human capital. Another feature of this Definitions of human capital have varied between an educated and study is that the density of human capital is included in the analysis, skilled workforce (Glaeser 1994, Glaeser & Saiz, 2004; Alquézar by measuring the number of both relative and absolute indicators. The Sabadie & Johansen, 2010) to talented and tolerant citizens (Florida, study demonstrates that the relationship between tolerance and inno- 2002). The tolerance aspect has received several criticisms (e.g. Clark, vation or human capital measures only holds for highly populated 2003; Glaeser, 2004) and has therefore been further redefined areas on the postal code level (Clark, 2003). Bringing together the bal- (Florida, 2012). The debate centres on whether firms locate in areas ance between share and volume, i.e. the relative and absolute level of with an educated and skilled workforce (Florida, 2002, 2012), or human capital, is also a new view point to studies concerning HMA. whether employees follow the firms (e.g. Scott, 2000, 2006). Earlier These have usually examined proportions, and therefore have conclud- studies have also recognised that human capital observed from sur- ed that detached housing areas with a high share (relative) of profes- rounding areas is more influential concerning innovation clustering sionals are more important than more densely populated areas than the area itself (Simonen & McCann, 2008). Therefore, the spatial in- hosting a higher absolute number of these professionals for innovation terdependence between human capital and innovative output is also driven development and economic growth (e.g. Vaattovaara, 1998; tested. This enables us to answer the question of whether the human Vilkama, Lönnqvist, Väliniemi-Laurson, & Tuominen, 2014; Kiuru, capital of a neighbouring area influences the innovative capacity of 2015). small observation units, such as postal codes. We recognise the difficulty of measuring the different ways of inno- Considering single variables, tertiary education is one of the most vativeness, creativity and skill or know-how. Patents are probably the important indicators associated with the human capital concept. most commonly used indicator (e.g. Carlino, Chatterjee, & Hunt, 2007; J. Kiuru, T. InkinenCities 64 (2017) 9–17 11

Kerr & Kominers, 2010; Carlino, Carr, Hunt, & Smith, 2012; Murata, number and share of professionals working in knowledge-intensive Nakajima, Okamoto, & Tamura, 2012), but patents have also faced crit- fields(e.g.R&D,education,ICT);thenumberandshareofartists; icism since they quantify mainly technical innovations, far too often ex- and the absolute and relative popularity of immigration-critical cluding social innovations, for example. In addition, several patents parties. This last point needs some additional explanation: both abso- never reach the market and their importance is therefore questionable. lute and relative votes were taken into consideration for the following R&D activity is another way of examining a region's innovativeness (e.g. parties in the 2011 parliamentary election: True Finns, the Indepen- Audretsch & Feldman, 1996; Kenney, 2000; Agrawal, Cockburn, & dence Party, and Change 2011, all of which had immigration criticism Rosell, 2010; Lee & Nicholas, 2012), but R&D projects do not necessarily on their agenda). turn into innovations, however. R&D expenditure is also an innovation The variables presented in Table 1 were treated as follows: R&D ac- input and that tells nothing about the outputs to or impacts on markets. tivity was measured with data from Tekes (The Finnish Funding Agen- One way is to examine the amount of knowledge intensive business ser- cy for Innovation). The data consisted of rows of granted R&D projects vice (KIBS) firms (e.g. Manniche, 2012; Inkinen & Kaakinen, 2016). In with information on the postal code area of the applicant and the addition, in the few decades since the work of Jacobs (1969),urbanden- amount of funding (in Euros) that was granted. We combined these sity has again been raised as an essential factor for innovative and eco- attributes with GIS data on postal code areas (from HSY) to conduct nomic growth (e.g. Fritsch, 2004; Boschma & Fritch, 2009; Malizia & a spatial analysis. Tekes is only one of the funding sources for R&D de- Motoyamab, 2015). Therefore, we will examine both the absolute and velopment, but we hypothesised that the funding from the most im- relative innovativeness of postal code areas. portant single innovation agency sufficiently indicated broader R&D One interesting aspect of earlier innovation studies is the discovery activity. In our analysis, we included the number of grants for the pri- of innovation paradoxes. Rodriguez-Pose (1999) discovered that some vate sector, as well as the total sum of money for the private sector. regions exhibit stronger (innovation averse) and some regions exhibit The data concerning patents was obtained from the Finnish Patent weaker (innovation prone) than expected economic growth relative and Register Office, and only patents that had been applied for by to their R&D activity (see also Makkonen & Inkinen, 2013). The same the private sector were included. Thus, commercial innovation clus- concept could be implemented in measuring the region's innovative tering could be examined. We converted the raw data into GIS data, growth regarding their human capital. as was done with the R&D projects. From these starting points, we examine the spatial relationship be- Information concerning tertiary degrees, students and residents tween an educated, skilled and tolerant workforce (human capital), aged 35–44 are all open data in the Statistics Finland website (data- and innovative output in intra-metropolitan clusters. As the causation base Paavo). The information is available at the postal code level, en- between human capital and innovations would need longitudinal abling us to examine the intra-metropolitan spatial distribution. The data, which we do not have, we are unable to say which came first: education data was combined with the GIS data, as was the case human capital (Florida, 2002) or innovations (Scott, 2000, 2006). This with the innovation data. Information on elections is classified with is mainly because the standard industrial classification in Finland respect to election areas, which are slightly different from postal changed in 2008, making the comparison between the years before code areas. This problem was solved by using the “proportion sum” and after that date very problematic, if not impossible. An educated, operation in MapInfo software, which placed the number of election skilled and tolerant workforce may generate innovation, but also area votes to postal codes regarding the proportion of spatial overlap- knowledge-intensive firms and their establishments may attract profes- ping. Other human capital variables were obtained from the Finnish sionals. In this respect, the last research problem was to identify clusters Environment Institute (YKR). This is the only data that has a charge- that underperform regarding the level of nearby human capital, and able license. The data consists of 250 m cells, which have multiple at- clusters that are more innovative than expected regarding the proxim- tributes related mainly to the urban and social structure. We ity of human capital. We predict potential clusters of innovations, as aggregated the cell data into postal code areas with respect to the pro- well as include the future demand for increasing the number of profes- portion of each cell overlapping the postal code area. The latest data of sionals. Findings of underperforming spatial areas can be implemented some of the indicators are from 2012, so we used that year in our other initially in HMA strategic planning. These areas could benefit, for exam- variables as well. ple, from commercial zoning, new master plans and actions from busi- After the data was collected, we applied Principal Component ness services departments. On the other hand, overachieving areas Analysis (PCA) to innovation indicators with IBM SPSS Statistics soft- could benefit from new residential development motivated by the in- ware. By doing that, we could find out which of the indicators were creased demand of the new professionals. significant for measuring innovativeness. With PCA scores of the most potent component we identified innovation clusters in the 3. Data and methods HMA with one measure. Clusters were determined simply by classify- ing postal code areas into three categories regarding their PCA scores. First, we collected potential indicators used to measure innovative- After establishing the innovation score, we tested all the potential ness in earlier literature (see upper part of the Table 1). After identifying human capital variables using OLS regression analysis (the only op- the most significant input and output innovation indicators, we also tion of statistical regression in GeoDa software). Because there was a considered the absolute and relative measures depicting innovation. relatively large number of variables (16, Table 1)comparedtothe After collecting statistical data, we obtained the number of KIBS estab- sample size (179 postal code areas), the variables were split into lishments (a point pattern GIS data) from the Helsinki Region Environ- two groups of predictive components. The logical division was to sep- mental Services Authority (HSY). Data is freely available for research arate the groups into absolute variables and relative indicators. The purposes. We included only private sector firms in our analysis, follow- constant variable was the PCA score of the innovation indicators. ing the classification defined by Inkinen and Kaakinen (2016) in their The literature concerning innovative capacity includes a conclu- earlier study. There are three occupational classes: Class I includes the sion that human capital from other regions has more influence on in- ICT sector establishments, Class II consists of R&D and education work- novation clustering than the human capital of the area itself (Simonen places, and Class III is formed from business services jobs. A fourth &McCann,2008). Therefore, the spatial correlation between human variable combines each subcategory into a sum variable (total number capital and innovativeness was tested using also spatial regression of class I-III establishments). analysis. Thus, we analysed whether or not the human capital of In the second phase, we collected potential indicators for neighbour areas influenced the innovative capacity. Similarly, in the measuring human capital (see the bottom part of Table 1). These in- field of urban geography, Song (2014) has used spatial regression in clude, for example, the number and share of tertiary degrees; the the analysis of land cover in Beijing. Similarly, Chi (2011) predicted 12 J. Kiuru, T. InkinenCities 64 (2017) 9–17

Table 1 Innovation and human capital indicators with descriptive statistics.

Variables for innovation Source Availability Min Max Median Average St. dev.

the absolute number of area's private sector knowledge intensive jobs Helsinki Region Environmental Free when 0 9486 89 524 1256 (inputs) Services Authority (HSY) asked the relative share of area's private sector knowledge intensive jobs of Helsinki Region Environmental Free when 073 121613 area's total jobs (inputs) Services Authority (HSY) asked the absolute number of area's private sector patents (outputs) Finnish Patent and Register Office Free when 066 0 2 8 asked the relative share of area's private sector patents of area's total jobs Finnish Patent and Register Office Free when 0 0.04 0 0.001 0.004 (outputs) asked the absolute number of area's private sector R&D projects The Finnish Funding Agency for Free when 0 69 0 0.3 8 Innovation (Tekes) asked the relative share of area's private sector R&D projects of area's total jobs The Finnish Funding Agency for Free when 0 0.02 0 0.001 0.002 (inputs) Innovation (Tekes) asked the absolute number (euros) of area's R&D spending (inputs) The Finnish Funding Agency for Free when 0 15,339,350 0 746,867 2,123,592 Innovation (Tekes) asked the relative share of area's R&D spending of area's total jobs (inputs) The Finnish Funding Agency for Free when 0 2676 0 200 456 Innovation (Tekes) asked

Variables for human capital Source Availability Min Max Median Average St. dev

The absolute number of area's tertiary degrees Statistics Finland (Database “Paavo”) Open data 0 4929 636 854 762 The relative number of tertiary degrees of area's total workforce Statistics Finland (Database “Paavo”) Open data 0 40 17 18 10 The absolute number of area's students Statistics Finland (Database “Paavo”) Open data 0 2169 400 463 363 The relative number of students of area's total workforce Statistics Finland (Database “Paavo”) Open data 0 37 7 8 4 The absolute number of 35–44-year-old residents Statistics Finland (Database “Paavo”) Open data 0 3085 727 843 645 The relative number of 35–44-year-old residents of area's total residents Statistics Finland (Database “Paavo”) Open data 0 21 14 14 4 The absolute number of professionals working in R&D Finnish Environment Institute Paid 0 6545 62 294 718 (Database “YKR”) The relative number of professionals working in R&D of area's total workforce Finnish Environment Institute Paid 0 48 6 8 7 (Database “YKR”) The absolute number of professionals working in the fields of the ICT industry Finnish Environment Institute Paid 0 4915 20 268 758 (Database “YKR”) The relative number of professionals working in the fields of the ICT industry of Finnish Environment Institute Paid 0 52 2 5 8 area's total workforce (Database “YKR”) The absolute number of professionals working in finance and insurance Finnish Environment Institute Paid 0 4546 3 130 495 (Database “YKR”) The relative number of professionals working in finance and insurance of area's Finnish Environment Institute Paid 0 30 0.4 2 4 total workforce (Database “YKR”) The absolute number of artists Finnish Environment Institute Paid 0 1760 20 82 183 (Database “YKR”) The relative number of artists of area's total workforce Finnish Environment Institute Paid 0 17 2 3 3 (Database “YKR”) The absolute number of votes for immigration-critical parties Finnish Environment Institute Paid 0.002 2992 375 532 506 (Database “YKR”) The relative number of votes for immigration-critical parties of area's total votes Finnish Environment Institute Paid 4.59 31 16 16 6 (Database “YKR”)

population growth in the census tracts of Milwaukee using spatial re- 4. Results gression, and Duncan (2013) used the method to assess the connec- tion between urban form and depression. 4.1. Principal component analysis, OLS and spatial lag models Similarly, in spatial lag regression analysis, the constant variable was the PCA score of the innovation indicators. The predictors were absolute The first component of the PCA analysis had an explanation level of indicators from Table 1 in the first analysis and relative indicators from 42% compared to 14% of the second component (Table 2). The first com- Table 1 in the second analysis. After two OLS analyses and two spatial ponent was dominated by absolute indicators, whereas the second lag analyses, a final OLS and spatial lag analyses (all six models in component consisted of relative indicators. We used the PCA score of Table 3) were made of all the significant variables from the first four the first component in further analysis. The relative number of patents analyses. This allowed us to measure human capital and predicted and relative number of R&D establishments were the only ones that growth with the combined model using both level and proportional were not significant in the first component, and therefore were left variables. out from the innovativeness score. With the predicted innovation values of the final analysis, we iden- After finding out the innovation score, we tested all the potential tified the clusters of human capital in the HMA. We classified postal human capital variables by using OLS regression analysis. Looking at code areas into three classes and presented them on two thematic the F-Statistic value, it is apparent that both tests reached the maximum maps (Figs. 1 and 2). Further, using spatial regression, we revealed the statistical significance (b0.001) values, respectively (Table 3,models1 correlation between nearby human capital and innovativeness, and pre- & 2). R-squared values indicate that the variables chosen to explain dicted the areas that could potentially be more innovative. We used re- the innovative output were highly effective, as the group of absolute in- sidual values, i.e. the gap between true and predicted innovative output, dicators explains 85% of the dependent variable and the group of pro- of the final spatial lag analysis. Conversely, with negative residual portional variables explains 44%. This result means that the indicators values, we predicted the areas that could potentially attract more pro- may be used in studies of other city regions as well. fessionals. We performed the analysis using GeoDa open source soft- Further analysis of the results of the regression analysis, revealed that ware. We illustrated the analyses with MapInfo software. independent variables were not highly correlated (the multicollinearity J. Kiuru, T. InkinenCities 64 (2017) 9–17 13

Fig. 2. Predicted geography of innovation prone and future demand areas in the HMA. value is 14 in an absolute dataset and 20 in a relative dataset). The next predictions. In fact, the model explained only 85% of innovativeness step was to find out whether there was a spatial dependence, i.e. that with the absolute variables. The model with proportional variables the values of the nearest neighbours were significant, between predictors functioned in the same way: explanation was lower with a contiguity and constant variables. Regarding spatial dependence, it is apparent that of two (45%) compared to a contiguity of one (49%). both lag and error Lagrange Multiplier (LM) values were significant in In the case of individual indicators (Table 1), the only variable that both datasets. Robust LM suggests that there was only a spatially lagged did not predict the innovative output of locations, in neither absolute dependence in both datasets. Therefore, a spatial lag analysis was con- nor relative terms, was the resident's age when approaching 40. All ducted for both absolute and relative variables. the other indicators had significance at least in one of the four examined From the spatial lag analysis, we observed that consideration of the models (Table 3, models 1, 2, 3 & 4). To measure total human capital and level of human capital of the neighbouring areas (in addition to the predicted growth with combined level and proportional variables, we area itself) added analytical value. The neighbourhood effect was evi- conducted a final analysis (Table 3, models 5 & 6), using only significant dent: the absolute level of human capital prediction was 88% of the in- variables from the first four regression analyses. We interpreted that novative capacity of an area and its neighbours, compared to 85% if scores of the first component of the PCA represented innovativeness. the predicted area was considered alone. In the case of relative human Predicted innovation scores of the final OLS analysis represented the el- capital, the corresponding number rose from 44% to 49% when emental variables from the human capital indicator set (see Table 1). neighbouring areas were considered. Adding the second order of conti- Further, we predicted areas defined as having innovation potential, guity (the neighbours' neighbour) gave no additional value to our and areas with future demand for skilled professionals with residual

Table 2 Principal component analysis results for original data (N =179).

Extraction method: Principal component analysis. 4 components extracted Component

1234

Absolute number (PCS.) of R&D projects 0.902 −0.237 0.062 −0.194 Absolute money (euros) spent on R&D 0.879 −0.215 0.092 −0.230 Absolute number of patents 0.499 −0.120 0.297 0.481 Relative number (PCS.) of R&D projects (compared to number of KIBS establishments) 0.349 0.704 0.368 −0.326 Relative money (EUROS) spent on R&D (compared to number of KIBS establishments) 0.381 0.609 0.287 −0.419 Relative number of patents (compared to number of KIBS establishments) 0.069 0.050 0.359 0.707 Absolute number of it establishments 0.865 −0.240 −0.016 0.015 Absolute number of R&D establishments 0.597 0.007 −0.609 0.024 Absolute number of business services establishments 0.844 −0.297 −0.033 −0.079 Absolute number of KIBS establishments altogether 0.942 −0.281 −0.082 −0.035 Relative number of it establishments (compared to number of establishments) 0.682 0.101 0.151 0.193 Relative number of R&D establishments (compared to number of establishments) 0.108 0.473 −0.759 0.138 Relative number of business services establishments (compared to number of establishments) 0.418 0.477 0.206 0.229 Relative number of KIBS establishments altogether (compared to number of establishments) 0.623 0.612 −0.277 0.306 Eigen value 5.86 2.037 1.54 1.311 % of variance 42% 14% 11% 9% 14 J. Kiuru, T. InkinenCities 64 (2017) 9–17

Table 3 OLS-models (1, 2 and 5) for absolute, relative and combined variables; and corresponding spatial lag models (3, 4 and 6). Dependent variable in all models is the innovation score.

Dependent variable: Innovation Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Score (OLS_absolute) (OLS_relative) (SpaLag_absolute) (SpaLag_relative) (OLS_combined) (SpaLag_combined)

Constant −0.385 −0.279 0.348 −0.431 −0.165 −0.226 t-stat or z-value (sig) −7.298 (***) −0.738 −7.554 (***) −1.213 −1.029 −1.526

Independent variables Lagged innovation score Excluded Excluded 0.292 0.294 Excluded 0.237 z-value (sig) 6.624 (***) 3.323 (***) 4.800 (***) Students_abs −2.320e-005 Excluded 3.185e-005 Excluded Excluded Excluded t-stat or z-value (sig) −0.118 0.185 Immigrant_abs −2.581e-005 Excluded 1.682e-005 Excluded Excluded Excluded t-stat or z-value (sig) −0.316 0.236 Age_abs 1.891e-005 Excluded −1.074e-005 Excluded Excluded Excluded t-stat or z-value (sig) −1.369 −0.899 Tertiary_abs 2.941e-005 Excluded 1.308e-005 Excluded 6.229e-005 3.903e-005 t-stat or z-value (sig) 3.741 (***) 1.818 1.281 0.879 IT_abs 5.720e-005 Excluded 5.579e-005 Excluded 4.411e-005 4.862e-005 t-stat or z-value (sig) 9.124 (***) 10.242 (***) 5.694261 (***) 6.817 (***) Finance_abs −1.275e-005 Excluded 3.646e-006 Excluded N/A N/A t-stat or z-value (sig) −0.140 0.046 R_D_abs 8.394e-005 Excluded 7.942e-005 Excluded 8.525e-005 8.264e-005 t-stat or z-value (sig) 10.738 (***) 11.552 (***) 10.436 (***) 11.023 (***) Arts_abs −11.652e-005 Excluded −12.369e-005 Excluded −9.852e-005 −11.611e-005 t-stat or z-value (sig) −4.201 (***) −5.127 (***) −3.850 (***) −4.930 (***) Students_prop Excluded 0.039 Excluded 0.036 −0.001 −0.005 t-stat or z-value (sig) 2.43 (*) 2.371 −0.135 −0.696 Immigrant_prop Excluded −0.041 Excluded −0.024 −0.019 −0.009 t-stat or z-value (sig) −2.692 (**) −1.593 −3.174 (**) −1.427 Age_prop Excluded 0.029 Excluded 0.027 Excluded Excluded t-stat or z-value (sig) 1.468 1.427 Tertiary_prop Excluded −0.013 Excluded −0.013 Excluded Excluded t-stat or z-value (sig) −1.271 −1.328 IT_prop Excluded 0.051 Excluded 0.046 0.012 0.007 t-stat or z-value (sig) 6.070 (***) 5.828 (***) 2.195 (*) 1.273 Finance_prop Excluded 0.042 Excluded 0.036 0.011 0.008 t-stat or z-value (sig) 2.931 (**) 2.680 (**) 1.465 1.136 R_D_prop Excluded 0.029 Excluded 0.024 0.006 0.004 t-stat or z-value (sig) 2.901 (**) 2.633 (**) 1.268 0.824 Arts_prop Excluded −0.027 Excluded −0.028 Excluded Excluded t-stat or z-value (sig) −1.569 −1.731

Model summaries Number of cases (N) 179 179 179 179 179 179 Number of variables 9 9 10 10 10 11 Degrees of freedom 170 170 169 169 169 168 R-squared 0.847 0.444 0.878 0.486 0.866 0.883 Adjusted R-squared 0.840 0.418 N/A N/A 0.859 N/A Sum squared residual 26.544 96.218 N/A N/A 23.164 N/A Sigma-square 0.156 0.566 0.118 0.497 0.137 0.1136 S.E. of regression 0.395 0.752 0.343 0.705 0.370 0.3370 F-statistic 117.348 16.985 N/A N/A 121.567 N/A Prob (F-statistic) 0.000 (***) 0.000 (***) N/A N/A 0.000 (***) N/A Log likelihood −83.171 −198.431 −63.961 −192.950 −70.981 −60.256 Akaike info criterion 184.342 414.862 147.921 405.900 161.962 142.511 Schwarz criterion 213.029 443.549 179.795 437.774 193.835 177.572

* = sig. b0.05 ** = sig. b0.01 *** = sig. b0.001 values from the final spatial lag analysis. All in all, there were five There were three indicators that predicted innovativeness in our human capital indicators that predicted postal code level innovative- combined spatial lag-model 6 (Table 3). Indicators that influence inno- ness in our final model combined OLS model 5 (Table 3): vativeness, including neighbouring areas, are: LIS = lagged innovation IS=0.441∗IT_abs+0.853∗R_D_abs−19.264∗Immigrant_prop+ score itself (z-value: 4.800; sig. 0.000; N = 179); IT_abs = The absolute 12.025∗IT_prop−0.985∗Arts_abs number of ICT professionals (z-value: 6.82; sig. 0.000; N = 179); The dependent variable (IS) is the calculated innovation score R_D_abs = The absolute number of R&D professionals (z-value: (*1000) and the explanative variables are: IT_abs = The absolute num- 11.02; sig. 0.000; N = 179); Arts_abs = the absolute number of artists ber of ICT professionals (t-statistic: 5.69; sig. 0.000; N = 179); (z-value: −4.93; sig. 0.000; N =179). R_D_abs = the absolute number of R&D professionals (t-statistic: As a summary of the results, it is easy to see that the same indicators 10.43; sig. 0.000; N = 179); Immigrant_prop = the relative popularity have the greatest significance regardless whether they are used in OLS of immigration criticism (t-statistic: −3.17; sig. 0.002; N = 179); or spatial lag models. Spatial lag models provided the highest explana- IT_prop = the proportion of ICT professionals (t-statistic: 2.20; sig. tive power (model 6 has the highest R-square value 0.882). The results 0.030; N = 179); and Arts_abs = the absolute number of artists (t-sta- indicate that areas benefit, in terms of innovation, if they are tolerant tistic: −3.85; sig. 0.000; N =179). (immigration criticism t-statistics are significantly negative in OLS J. Kiuru, T. InkinenCities 64 (2017) 9–17 15 models 2 and 5). This is not a surprising result compared to the signifi- Table 5 cant negative t-statistics of the absolute number of artists in models 1, 3, Top 15 predicted innovation and future potential demand postal code areas in the HMA. 5 and 6, in which the variable was included. Area Innovation Area Potential The results show that spatial autocorrelation exists to some extent in potential demand the HMA. This corresponds with the conclusion that human capital in Länsi- −1.8009 Pitäjänmäen teollisuus 1.2273 neighbouring areas is important for growth (Simonen & McCann, −1.4344 Pohjois-Tapiola 1.0421 2008). Adding one layer of neighbouring areas contributed an addition- −0.6989 1.0241 − al 9% of explanative power to the models. Adding the second contiguity Laajalahti-Friisinmä 0.6776 Pohjois-Leppävaara 0.9106 − fi Jätkäsaari 0.6535 Meilahden sairaala-alue 0.8069 layer had no signi cant effect on the results. Instead, explanation levels Helsinki Keskusta −0.5592 Westend 0.7173 dropped approximately 5% from the highest levels. −0.5436 0.6575 −0.5292 Nihtisilta 0.6325 4.2. Spatial distributions, area ranks and indicator scores Vattuniemi −0.4615 Pajamäki 0.5893 Jokiniemi −0.4333 Oitmäki 0.5782 −0.4301 Vantaanpuisto 0.4788 Our research questions are addressed and answered in this spatial Puolarmetsän sairaala −0.4051 Kirkonkylä-Veromäki 0.4562 analysis. Firstly, the studied innovation clusters and neighbourhoods Espoon Keskus länsi −0.3516 Tapiola 0.4559 (postal code areas) with the highest level of human capital are illustrat- Kuurinniitty −0.3456 Hämevaara 0.4299 − ed in Fig. 1. It is evident that concentrations of innovations are highly 0.3295 0.4291 correlated with concentrations of human capital. Further, it is evident that innovations and human capital are highly clustered (see combined scores in Table 4). Spatially, one larger cluster is located in Western Hel- businesses, in relation to the persons living in those areas. Potential de- sinki and Eastern Espoo. We called this cluster an innovation horseshoe. mand refers to areas that experience more highly skilled residents (high Secondly, the predicted postal code areas for potential innovation loca- human capital) than the actual presence of knowledge-intensive firms tions and professional demand locations are presented in Fig. 2 and would predict. This is an important observation, as it enables the map- Table 5. Our cluster identification corresponds with earlier HMA innova- ping exercise presented in the Fig. 2. tion studies (e.g. Inkinen & Kaakinen, 2016), but differentiates largely The final interpretation concerning Tables 4 and 5 is that the relative from earlier HMA housing studies, which have seen low density resi- distance in the combined performance of innovation and human capital dential areas as the areas with the highest socioeconomic status (e.g. is dualistic. Thus, there are two clear groups identifi able in Table 4. First, Vaattovaara, 1998; Vilkama et al., 2014; Kiuru, 2015). Agglomeration the top three received a combined score close to a value of 10. All the of human capital externalities in the most central parts of metropolitan other top areas had significantly lower scores and their values were area is also in line with the concept of Smart Café Cities (Fu, 2007). close to each other. In other words, their relative distances are short in Fig. 1 and Table 4 indicate both the geographical locations and statis- our variable metrics. We can interpret this result as empirical evidence tical values of the most innovative and human capital intensive loca- of a strong clustering tendency concerning innovation and human cap- tions within the HMA. The top three locations in both categories are ital: it is likely that cities are able to support a limited number of high- the Helsinki centre, Ruoholahti and Otaniemi (Inkinen, 2015). Also on end hotspots in this field. We consider this result to be valid at least the broader geographical scale, nationally these three postal code for cities close to the size of HMA (population approx. 1 to 1.5 million). areas present the most prominent locations of innovation and human capital in Finland. Fig. 1 highlights the co-location of innovation and 5. Discussion and conclusions human capital. Geographical contrast is significant if these results are compared to the Fig. 2 and Table 5 that present potential growth The results of this study contribute to the international debate deal- areas. The division into Northern and Southern parts and the horseshoe ing with the presence of human capital in innovation clusters. First, the development, following the main ring roads, is clearly visible. A compar- results indicate that human capital-driven innovation clustering is a ison between the categories of Table 5 indicates that different locations localised intra-metropolitan phenomenon. In general, knowledge- are leading their respective fields with dispersed geographies. We inter- intensive firms locate in proximity to highly skilled workforces within pret this to be an indication of the functionality of our division into cur- the HMA. The second conclusion is that the measurement of innova- rent leading innovation hotspots and potential future demand areas. tions should include a rigid distinction between relative and absolute Negative values on Table 5 concerning innovation potential are variables with the consideration of innovation input and output proxies. interpreted as the potential between the observed spatially lagged em- Similarly, in the prediction of locational innovativeness, relative and ab- ployment opportunities, thus the presence of knowledge-intensive solute indicators of human capital are also important. Interestingly,

Table 4 Highest 15 ranking postal code locations in the HMA in terms of combined innovation and human capital scores.

Combined area rank Total score Innovation rank Innovation score Human capital rank Human capital score

Helsinki Center 11.93 Helsinki Center 5.57 Helsinki Center 6.35 Ruoholahti 10.32 Ruoholahti 5.56 Ruoholahti 4.76 Otaniemi 9.40 Otaniemi 4.83 Otaniemi 4.56 Pitäjänmäen teollisuus 6.46 Pitäjänmäen teollisuus 3.89 Etelä-Leppävaara 2.64 Etelä-Leppävaara 5.51 Etelä-Leppävaara 2.87 Pitäjänmäen teollisuus 2.56 Pohjois-Tapiola 3.79 Pohjois-Tapiola 2.46 Länsi-Pasila 2.24 Etu- 3.61 Itä-Pasila 1.91 Eira 1.81 Itä-Pasila 3.50 Etu-Vallila 1.91 Vattuniemi 1.72 Punavuori 3.20 1.75 Punavuori 1.72 Kaartinkaupunki 3.18 Munkkiniemi 1.56 Etu-Vallila 1.70 Vattuniemi 3.17 Nihtisilta 1.52 Itä-Pasila 1.59 Eira 2.82 Punavuori 1.48 Kaartinkaupunki 1.43 Niittykumpu 2.62 Vattuniemi 1.45 Pohjois-Tapiola 1.33 Nihtisilta 2.40 Niittykumpu 1.44 Pikku Huopalahti 1.24 Munkkiniemi 2.16 Eira 1.01 Niittykumpu 1.19 16 J. Kiuru, T. InkinenCities 64 (2017) 9–17 often-used indicators such as the relative number of tertiary degrees, Boschma, R. A., & Fritch, M. (2009). Creative class and regional growth: Empirical evi- – – dence from seven European countries. Economic Geography, 85(4), 391 423. the absolute and relative number of citizens aged 35 44, and the rela- Carlino, G., Chatterjee, S., & Hunt, R. (2007). Urban density and the rate of invention. tive number of artists, are not statistically significant factors in any of Journal of Urban Economics, 61,389–419. our models. In fact, the absolute number of artists correlated negatively Carlino, G., Carr, J., Hunt, R., & Smith, T. (2012). The agglomeration of R&D labs. (Working Paper). with innovativeness. The number of artists may prove to be more signif- Chi, G. (2011). Small-area population forecasting in an urban setting: A spatial regression icant if broader spatial categories are applied, but as the analysis has approach. Journal of Population Research, 28(2–3), 185–201. shown, their presence does not contribute to clustering on a small Clark, T. N. (2003). Urban amenities: Lakes, opera and juice bars, do they drive develop- scale such as postal codes. ment? In T. N. Clark (Ed.), The city as an entertainment machine. Research in urban policy, Vol. 9.(pp.103–140). Oxford: Elsevier Ltd. In addition to traditional innovation and human capital indicators Doms, M., Lewis, E., & Robb, A. (2010). Local labor force education, new business charac- obtained from earlier literature, a new indicator was collected and ap- teristics, and firm performance. Journal of Urban Economics, 67(1), 61–77. plied, namely the popularity of immigration-critical parties. The reason Duncan, D. T. (2013). The built environment and depressive symptoms among urban youth: A spatial regression study. Spatial and Spatio-Temporal Epidemiology, 5, for this was to establish an insight into the spirit of innovative tolerance. 11–25. Popularity of the immigration-critical parties proved to measure areal Florida, R. (2002). The rise of the creative class. New York: Basic Books. tolerance as the proportion of votes for immigration-critical parties cor- Florida, R. (2012). The rise of the creative class: Revisited. New York: Basic Books. Florida, R., Mellander, C., & Stolarick, K. (2008). Inside the black box of regional develop- related negatively with innovativeness. We also wanted to use the num- ment. Human capital, the creative class and tolerance. Journal of Economic Geography, ber and proportion of students involved, which may be considered an 8,615–649. unconventional measure. Fritsch, M. (2004). Entrepreneurship, entry and performance of new business compared fi in two growth regimes: East and West Germany. Journal of Evolutionary Economics, The empirical results illustrate that absolute variables are signi - 14(5), 525–542. cantly more effective than relative indicators. The dominance of abso- Fu, S. (2007). Smart café cities: Testing human capital externalities in the Boston metro- lute indicators in both innovativeness and human capital suggest that politan area. Journal of Urban Economics, 61,86–111. Glaeser, E. L., & Feldman, E. L. M. P. (2000). The new economics of urban and regional urban density is an essential, and often underrated, circumstance for in- growth. In G. L. Clark, & M. S. Gertler (Eds.), The Oxford handbook of economic novative growth. Considering planning and the mixed land use para- geography. digm, the results show clearly that innovations emerge the best in Glaeser, E. L. (2004). Book review of Richard Florida's “The Rise of the Creative Class”. dense and mixed urban structure. Glaeser, E. L., & Kerr, W. R. (2009). Local industrial conditions and entrepreneurship: How much of the spatial distribution can we explain. NBER working paper no. 14407. The analysis identified biases in the balance between housing and Glaeser, E. L., & Saiz, A. (2004). The rise of the skilled city. Brookings Wharton papers on knowledge-intensive establishments. Thus, innovation prone and inno- urban affairs. 5.(pp.47–94). vation averse postal codes were interpreted as being areas with innova- Glaeser, E. L., Kerr, W., & Ponzetto, G. (2010). Clusters of entrepreneurship. Journal of Urban Economics, 67(1), 150–168. tion potential and areas with future demand for skilled professionals. Henry, N., & Pinch, S. (2000). Spatialising knowledge: Placing the knowledge community Clusters of innovation and human capital, and clusters with potential of Motor Sport Valley. Geoforum, 31(2), 191–208. growth characteristically form a larger spatial entity (an innovation Inkinen, T. (2015). Reflections on the innovative city: Examining three innovative loca- – fi tions in a knowledge bases framework. Journal of Open Innovation, 1(8), 1 23. horseshoe). This nding is in line with the Smart Café City concept Inkinen, T., & Kaakinen, I. (2016). Economic geography of knowledge intensive technolo- (Fu, 2007), where human capital externalities are highly localised in gy clusters: Lessons from the Helsinki metropolitan area. Journal of Urban Technology, the most central areas of the metropolitan areas. 23(1), 95–114. Jacobs, J. (1969). The economies of cities. New York: Random House. The city of Helsinki is currently planning several inner-city exten- Kenney, M. (2000). Understanding Silicon Valley: The anatomy of an entrepreneurial region, sions, and this analysis indicates that they could start from Southern, Palo Alto, CA. Stanford University Press. Eastern and Northwest locations, which could benefit from the zoning Kerr, W., & Kominers, S. (2010). Agglomerative forces and cluster shapes. NBER working paper 16639. of commercial space. Housing development could succeed particularly Kiuru, J. (2015). Pääkaupunkiseudun hyvä- ja huono-osaiset klusterit ja niiden muutos on the Western shore of the inner city (Ruoholahti to Pitäjänmäki). 2007–2012. Finnish Journal of Urban Studies, 54(4) Full text available at (Retrieved The most promising part of the surrounding city of Espoo is the Eastern 11.12.2016) http://www.yss.fi/journal/paakaupunkiseudun-hyva-ja-huono-osaiset- part of the city. These locations are also tightly connected to the centre klusterit-ja-niiden-muutos-2007-2012 Lawton Smith, H. (2009). Skill shortages, demographic aging, and training implications of Helsinki via a metro line, a commuter train line and an extensive road for skill-based economies. The Professional Geographer, 62(1), 59–69. infrastructure (ring roads one and two). According to our results, these Lee, J., & Nicholas, T. (2012). The origins and development of Silicon Valley. Harvard Busi- areas could attract additional professionals. Also in the Western part of ness School Case, 813-098. Makkonen, T., & Inkinen, T. (2013). Innovative capacity, educational attainment and eco- the city of Espoo, there is a cluster for potential housing. The third large nomic development in the European Union: Causal relations and geographical varia- city belonging to the HMA (Vantaa) has three larger clusters for poten- tions. European Planning Studies, 21(12), 1958–1976. tial growth: a postal code close to the International Airport, which could Makkonen, T., & Inkinen, T. (2015). Geographical and temporal variation of regional de- fi velopment and innovation in Finland. Fennia - International Journal of Geography, bene t from housing development; and Eastern and Western clusters, 193(1), 134–147. which could benefit from the zoning of commercial space. Development Malizia, E., & Motoyamab, Y. (2015). The economic development–vibrant center connec- actions from business services officials could also help these emerging tion: Tracking high-growth firms in the DC region. The Professional Geographer(Oct. 2015). locations to grow. Manniche, J. (2012). Combinatorial knowledge dynamics: On the usefulness of the dif- ferentiated knowledge bases model. European Planning Studies, 20(11), – Acknowledgements 1823 1841. Marlet, G., & Van Woerkens, C. (2004). Skills and creativity in a cross-section of Dutch cit- ies. Discussion paper series 04–29. Tjalling: C. Koopmans Research Institute. This research was funded by the Helsinki Metropolitan Region Urban Mellander, C., & Florida, R. (2006). The creative class or human capital? Explaining regional Research Program. development in Sweden. (Working paper). Murata, Y., Nakajima, R., Okamoto, R., & Tamura, R. (2012). Localized knowledge spillovers and patent citations: A distance-based approach. (Working Paper). References Porter, M. (2000). Location, competition, and economic development: Local clusters in a global economy. Economic Development Quarterly, 14(1), 15–34. Agrawal, A., Cockburn, I., & Rosell, C. (2010). Not invented here? Innovation in company Rauch, J. E. (1993). Productivity gains from geographic concentration of human capital: towns. Journal of Urban Economics, 67(1), 78–89. Evidence from the cities. Journal of Urban Economics, 34,380–400. Alquézar Sabadie, J., & Johansen, J. (2010). How do national economic competitiveness in- Rodriguez-Pose, A. (1999). Innovation prone and innovation averse societies: Economic dices view human capital? European Journal of Education, 45(2), 236–258. performance in Europe. Growth and Change, 30(1), 75–105. Audretsch, D., & Feldman, M. (1996). R&D spillovers and the geography of innovation and Scott, A. J. (2000). Global City-regions: Trends, theory, policy. Oxford: Oxford University production. American Economic Review, 86,630–640. Press. Barro, R. J. (1991). Economic growth in a cross section of countries. The Quarterly Journal Scott, A. J. (2006). Creative cities: Conceptual issues and policy questions. Journal of Urban of Economics, 106,407–443. Affairs, 28(1), 1–17. Bönte, W., Falck, O., & Heblich, S. (2009). The impact of regional age structure on entre- Simon, C. (1998). Human capital and metropolitan employment growth. Journal of Urban preneurship. Economic Geography, 85,269–287. Economics, 43,223–243. J. Kiuru, T. InkinenCities 64 (2017) 9–17 17

Simon, C., & Nardinelli, C. (1996). The talk of the town: Human capital, information and Vaattovaara, M. (1998). Pääkaupunkiseudun sosiaalinen erilaistuminen. Ympäristö ja the growth of English cities, 1861–1961. Explorations in Economic History, 33(3), alueellisuus. Helsingin kaupungin tietokeskus. Tutkimuksia, 1998, 7. 384–413. Vilkama, K., Lönnqvist, H., Väliniemi-Laurson, J., & Tuominen, M. (2014). Erilaistuva Simon, C., & Nardinelli, C. (2002). Human capital and the rise of American cities, pääkaupunkiseutu. Sosioekonomiset erot alueittain 2002–2012. Helsingin kaupungin 1900–1990. Regional Science and Urban Economics, 32(1), 59–96. tietokeskus. Tutkimuksia, 20141.. Simonen, J., & McCann, P. (2008). Firm innovation: The influence of R&D cooperation and World Economic Forum (2015). Global competitiveness report. Full report available at the geography of human capital inputs. Journal of Urban Economics, 64(1), 146–154. (Retrieved 14.10.2016) http://reports.weforum.org/global-competitiveness-report- Song, J. (2014). The relationships between landscape compositions and land surface tem- 2015-2016 perature: Quantifying their resolution sensitivity with spatial regression models. Zucker, L., Darby, M., & Brewer, M. (1997). Intellectual human capital and the birth of US Landscape and Urban Planning, 123,145–157. biotechnology enterprises. American Economic Review, 87,290–306. Stephan, P. E., & Levin, G. L. (2001). Exceptional contributions to US science by the foreign-born and foreign-educated. Population Research and Policy Review, 20(1), 59–79.